Human-in-the-loop agent for product regulatory screening: Case study on concrete

Current Automated Compliance Checking (ACC) research predominantly focuses on geometric auditing and system-level verification, leaving a theoretical gap in early-stage, specification-level checking of commercial product requirements derived from unstructured documents. Existing frameworks lack mechanisms to filter product alternatives based on text-derived technical criteria early in the specification workflow. To address this, this paper presents a human-in-the-loop agent that extends ACC beyond geometry into product regulatory screening. The system employs Large Language Model (LLM)-driven parsing and structuring to ingest diverse documents (e.g., EPDs, technical data sheets, drawings) into a unified schema, executing deterministic checks against parameterised regulatory codes. Validated through a case study on concrete using EN 206 and AS 3600, the approach demonstrates how formalising text-based specifications into machine-executable rules bridges the gap between compliance auditing and product selection. This provides a blueprint for transparent, scalable regulatory filtering across jurisdictions, enabling robust decision-support in procurement. • Expansion of Automated Compliance Checking beyond geometry to product specifications. • Hybrid framework combines LLM-driven parsing with deterministic rule-based logic. • Human-in-the-loop agent architecture validates inputs to mitigate AI hallucinations. • Automated pre-screening filters large product portfolios against regional regulatory standards. • Technical specifications, EPDs, and project documentation transformed into structured data.